Transcript Slide 1

Energy Efficient Ethernet
An Overview
Mike Bennett
Lawrence Berkeley National Lab
Agenda
• What is Energy Efficient Ethernet (EEE)?
• How we got this far
• Progress
• What’s next?
• Ethernet Alliance as an incubator for EEE
What is EEE?
• An idea
– Reduce energy consumption during periods of low
link utilization
• Low hanging fruit
• Minimizes impact on industry
• Minimizes impact on standard
• An IEEE 802.3 Study Group
– Formed in November 2006 to study the idea
• Technical and economic feasibility
• Compatibility and distinct identity
• Broad market potential
Things to be studied
• Rapid PHY Selection
• Switch to a lower speed (power) PHY during
low link utilization
– How to minimize transition time?
– How to avoid thrashing between speeds?
– Interaction with higher-layer protocols
– Link utilization on servers
– Interaction with control policy
• But, how did we get to this point?
The road to EEE
• Tutorial given in July 2005 to IEEE 802
– Bruce Nordman and Ken Christensen discuss the
problem and possible ways to solve it
• They propose the idea of Adaptive Link Rate
(ALR)
– This includes a control policy
• This raises awareness within IEEE 802
– Gauge interest between July ’05 and July ’06
• ALR whitepaper posted on EA site August ’06
– http://www.ethernetalliance.org/technology/white_papers/alr_v10.pdf
• CFI team forms in August ’06
The CFI – identify the problem
• “Big IT” – all electronics
– PCs/etc., consumer electronics, telephony
• Residential, commercial, industrial
– 200 TWh/year
– $16 billion/year
– Nearly 150 million tons
of CO2 per year
Numbers represent
U.S. only
One central baseload
power plant
(about 7 TWh/yr)
• Roughly equivalent to 30 million cars!
PCs etc. are digitally
networked now — Consumer
Electronics (CE) will be soon
The CFI – more about the problem
“ Little IT” — office equipment, network equipment,
servers
97 TWh/year
3% of national electricity
9% of commercial building electricity
Almost $8 billion/year
TWh/year
Network Equipment 13
Servers
PCs / Workstations 20
Imaging (Printers, etc.)
Monitors / Displays
UPS / Other
16
12
15
22
… However
Old data (energy use has risen)
Doesn’t include residential IT or networked CE
products
Note: Year 2000 data taken from Energy Consumption by Office and Telecommunications Equipment in Commercial Buildings--Volume
I: Energy Consumption Baseline Roth et al., 2002 Available at: http://www.eren.doe.gov/buildings/documents
The CFI – even more about the problem
Unrestrained
IT power
consumption
could eclipse
hardware
costs and put
great
pressure on
affordability,
data center
infrastructure,
and the
environment.
Source: Luiz André Barroso, (Google) “The Price of Performance,” ACM Queue, Vol. 2, No. 7, pp. 48-53, September 2005.
(Modified with permission.)
The CFI - motivation to start EEE
Electric Utility Rebates
• Appliances …
• HVAC systems …
• Lighting …
• … PC power supplies (2005)
• … Server computers (2006)
• …
Reference: http://www.pge.com/docs/pdfs/biz/rebates/hightech/DataCenters_slides.pdf
PG&E (California) provides rebates for more energy-efficient servers
EEE – motivation
• April 19, 2006 “Green Grid” formed
• “A group of technology industry leaders form The
Green Grid to help reduce growing power and
cooling demands in enterprise datacenters
• Energy Efficiency gets U.S. congressional
recognition
– July 12, 2006 House Resolution 5646 passes
• “To study and promote the use of energy efficient
computer servers in the United States”
• Green500 Supercomputer list created Nov. ‘06
– To “take energy efficiency into account when
ranking supercomputers”
• Using Flops/watt (total system power) as the metric
EEE – motivation
• Energy Star
– Requirements coming in 2009
• “All computers shall reduce their network link speeds
during times of low data traffic levels in accordance with
any industry standards that provide for quick transitions
among link rates”
• The market for Energy Efficient Ethernet
– Driven by customer’s desire to save energy costs
– Ethernet is used in markets where saving energy is crucial
– Accelerate deployment for new applications
– Enables use of incentives by energy industry
• Ultimately these translate to increased demand
The CFI – possible energy savings
Results from (rough) measurements
— all incremental AC power
— measuring 1st order
• Typical switch with 24
ports 10/100/1000
Mb/s
• Various computer
NICs averaged
The CFI – possible solution
• Desktop-to-switch links
– Are mostly idle
• Lots of very low bandwidth “chatter”
– High bandwidth needed for bursts
• Bursts are often seconds to hours apart
• Server links are also often not fully utilized
– Higher speed links offer more opportunity to save energy
– This is an area where more data is needed
• Evidence of low utilization (desktop users)
– LAN link utilization is generally in range 1 to 5% [1, 2]
– Utilization for “busiest” user in USF was 4% of 100 Mb/s
[1]
[2]
A. Odlyzko, “Data Networks are Lightly Utilized, and Will Stay That Way”, Review of Network Economics, Vol. 2, No. 3, pp. 210237, September 2003.
R. Pang, M. Allman, M. Bennett, J. Lee, V. Paxson, and B. Tierney, “A First Look at Modern Enterprise Traffic,” Proceedings of IMC
2005, October 2005
The CFI – possible solution
• Snapshot of a typical 100 Mb FE link
– Shows time versus utilization (trace from Portland State
Univ.)
Typical bursty usage
(utilization = 1.0 %)
100%
Utilization
80%
60%
40%
20%
0%
0
1000 2000 3000 4000 5000 6000 7000
Time (s)
The CFI – possible solution
• Control policy determines when to
transition data rate
– RPS can support many different control policies
– Need to consider but not define
• Trade-off of energy saved versus packet delay
– Energy savings achieved by operating at low data rate
– Delay occurs during transition from low to high data rate
Must prevent link from thrashing
Control policy runs
on both ends of the link
switch
link
desktop PC
Control policy can be
based on queue length
thresholds
The CFI – possible solution
Sequence
Device A
Device B
inhibit frame transmission
inhibit frame transmission
change link speed
change link speed
~1 ms
establish new data rate
allow frame transmission
allow frame transmission
Time
The CFI – possible solution
• Snapshot of a typical Ethernet link
Low rate time = 96.8%
with simulated RPS
Utilization
Mean packet delay with RPS is 0.67 ms
100%
Mean packet delay without RPS is 0.12 ms
Utilization-threshold policy is used
80%
High rate time = 3.2%
60%
40%
20%
0%
0
1000 2000 3000 4000 5000 6000 7000
Switching time:
Time (s)
1ms
Data rates: 100 and 10 Mb/s
Low/High thresh: 0KB and 32KB
The CFI – Benefits of EEE
• 1 Gb/s
– Most NICs and most energy to be saved
– Substantial benefits for homes and offices
– Battery life benefit for notebooks
• 10 Gb/s (copper)
– Reduces power burden in data centers
– Reduces cooling burden in data centers
– May increase switch/router port capacity
• Generally…
– Provides real economic benefit through energy savings
The CFI – potential savings identified
Assume 100% adoption (U.S. Only)
• Residential
– PCs, network equipment, other
– 1.73 to 2.60 TWh/year
– $139 to $208 million/year
• Commercial (Office)
– PCs, switches, printers, etc.
– 1.47 to 2.21 TWh/year
– $118 to $177 million/year
These figures do
not include
savings from
cooling/power
infrastructure
• Data Centers
– Servers, storage, switches, routers, etc.
– 0.53 to 1.05 TWh/year
– $42 to $84 million/year
Total: $298 to $469 million/year
EEE - progress
• November Plenary Meeting (CFI)
– 33 supporters
• 15 from EA member companies
– 73 attendees
• Passed 65 – 2 – 6
– Many EA members in the room
• January Interim Meeting
– Met for almost 2 days
• 27-30 people attended per day
– most were EA members
• Heard several presentations
– Update on control policies and impact to higher layer
protocols
– Open questions to the study group regarding scope
EEE - progress
• January Interim Meeting
– Presentations (continued)
• “Strawman” proposal for 5 criteria and objectives
• Update on candidate protocols
• Impact on clause 28 (auto-negotiation)
– Defined some terms
– Took several straw polls
– People took action items
• What’s next?
EEE - what’s next?
• Preparing for March
– Several people have offered to present
•
•
•
•
•
More simulations on control policy
Technical feasibility of Rapid PHY selection
Alternatives to Rapid PHY selection
Work related to Clause 28 (auto-negotiation)
A look at RPS for backplanes
EEE - what’s next?
• Preparing for March
– What we still need
• Traces from networks in different markets
– Server network utilization
– Home network utilization
– Consumer electronics network utilization
• How to communicate with upper layers?
– Don’t want a network melt-down because of RPS
• Economic feasibility of building EEE PHYs
• Further study on broad market potential
• Ultimately preparing for PAR
• Where does the Ethernet Alliance fit in this?
EEE/EA – Technology Incubation
• How did EEE get this far?
– ALR Whitepaper posted on EA site
– CFI Team and supporters
• Mostly EA members
– First study group meeting contributors and participants
• Mostly EA members
• EA Mission
– The mission is to promote industry awareness,
acceptance, and advancement of technology and
products based on both existing and emerging IEEE
802 Ethernet standards and their management.
EEE/EA – Technology Incubation
• Marketing Activities
– Proactive cohesive messaging to external
target audiences
• Press
– Most recently interviewed with CMP/Information Week to promote
EEE
» Arranged by the EA
• Planning to form an EEE technical subcommittee
– When the time is right
• Bottom line
– The EA has been there to help us be successful
• Membership has its benefits
EEE/EA – summary
• EEE is a classic case of technology
incubation by the EA
– EA support helped to keep EEE moving in the right
direction
• CFI preparation
– EA marketing support is raising industry awareness
• Press
• The Ethernet Alliance will be there for us
– When we need a place for consensus building among
members
– When the time comes for testing and demonstrating
interoperability
Thanks!
• Questions or comments?
– [email protected]